GPU Architecture

There are several differences between RDNA 2 and RDNA 3. However, the RX 7000 series has 20% more Compute Units (CUs), doubling the number of FP32 (32-bit floating point) computing functionality per CU. AMD still employs 64 Streaming Processors (SPs) per CU; however, that amount is more like double in many respects. Each GPU holds a single GCD chiplet (Graphics Compute Die) and up to six MCD chiplets (Memory Cache Die). The XTX has a fully enabled GCD and six MCDs.

In contrast, the 7900 XT has a partially enabled GCD with 84 CUs and five MCDs — the sixth MCD slot will remain occupied, either with a dummy piece of silicon or a non-functional MCD (non-functional because it is fused off, though it may or may not be defective). Transistor counts have increased significantly compared to the previous generation RX 6000-series, with 58 billion transistors compared to only 26.8 billion on Navi 21. Theoretical computation has increased by up to 159% compared to the RX 6950 XT, with 61.4 teraflops on the XTX card and 51.6 teraflops on the XT variant.